fuse |ai Fusemachines a Fellowship-2023 Wm Cuch iecicuikccumc ec melar itl) create advanced Al career opportunities SSFCourses: Microdegree™ in Al © Machine Learning (s months) + Deep Learning @ Months) Microdegree Specialization * Computer Vision (5 Months) * Natural Language Processing Months) ae Students are recommended to take the foundations course in programming and have got introduced to the basic idea of machine learning before joining this course. It is highly recommended that students have completed the following course or topics before beginning with this course Computer Science for Al + Python Programming and Common Python Libraries + Data Structures and Algorithms + REST API Mathematics for Al + Linear Algebra + Multivariate Calculus + Probability and Statistics Course Prerequisites:fuse (ai Machine Learning Course Objectiv: After the completion of this course, students will be able to: Explore the key concepts of machine learning workflow, supervised, Pia fark alee ele RUC loll machine learning. Summarize the intuition of parameter estimation techniques, aCe Cinta eaC he Cran Explain the idea behind decision trees, decision tree inducers and overfitting in decision trees. Apply KNN, SVM, distance-based partitioning algorithms and Le eee eae Ele e oma mL Ce Relate simple time series models with mixed and advanced models.fuse (ai Syllabus Module 1 Introduction to Machine Learning Introduction to Machine Learning, Data cleaning, Data Quality inspection and cleaning, Deduplication and missing value, Outlier handling and ML Workflow Module 2 _Linear Regression Introduction, Least Squares- Simple Regression, Multiple Regression and Geometric intuition, Gradient descent- Gradient descent method, Gradient descent for linear regression and Importance of Standardization, Regression Validation and Regularization, Ridge regression, Lasso regression and Elastic Net regression Module 3 Decision Trees Overfitting and Underfitting , Early Stopping, Pruning and importance, Impurity ‘metrics, Decision tree inducers,Decision tree implementation, Multivariate decision tree and Metrics and scoring Module 4 —_ Ensemble methods Error Decomposition, Ensemble, Bagging, Random Forest, Boosting, Adaboost, Gradient Boosting and XGBoost Module 5 Probabilistic Models Introduction to Probabilistic Models, Frequentist vs Bayesian Approach to Probabilistic Models, Introduction to probability distribution - Discrete probability distribution and Continuous probability distribution, Maximum Likelihood Estimation(MLE) for Multivariable Gaussian, Linear Regression and Logistic Regression, Maximum a Posteriori (MAP) and Bayesian Approach for Parameter estimation Module6 —Clustering Partition based clustering- KMeans Formal algorithm (Loss function non-convex) and KMeans variants, Hierarchical Clustering - Agglomerative clustering algorithm and BIRCH (Micro Phase: Macro Phase, Density Based Clustering- DBSCAN and OPTICS, Cluster ValidationModule 7 Module 8 Module 9 Syllabus Dimensionality Reduction Dimensionality reduction and SVD- Singular Value Decomposition (SVD) as a low- rank decomposition for any matrices and curse of dimensionality and necessity for dimensionality reduction, Principal Component Analysis(PCA)- objective and the purpose of PCA, working of PCA and the role of covariance matrix, eigenvalues and eigenvectors in it and application and limitation behind. Time Series Analysis Introduction to Time Series- Understand the concept of time-series and its difference from conventional machine learning problems, Naive TSA Models, Simple models- Autoregressive(AR) Models and Moving Average(MA) Models, Mixed models- ARMA model, ARIMA model, SARIMA model and ARIMAX and SARIMAX model, Exponential smoothing Reinforcement Learning Introduction to RL, Fundamental of Reinforcement Learning- K-arm bandits, Exploration vs Exploitation tradeoff, Markov Processes and Markov Reward Processes, Markov Decision Processes, Dynamic Programming, Sample based learning methods- Monte Carlo Methods, Temporal Differencefuse (ai Deep Learning Course Objective: Nm rofl) le Romeo Le Loc Bole Co old * Understand different types of Deep Learning Architectures, such as Oe aE RS elt om celled ULE Ceca ol CLAS Networks and Autoencoders. * Master the process of creating a Deep Learning pipeline OMT Ane OE LE RL Cece Rick kaa} models, validating your models and including regularization, Cu ke cL CRE RT eMch rch ccsfuse |ai Syllabus Neural Networks Introduction to Module, Computational Graphs and Backpropagation, Tensorflow ANN, Activation Function- sigmoid, tanh and also variations of ReLU, Weight Initialisation and types, Xavier Initialization and He and LeCun initialization, Optimization Methods- Stochastic Gradient Descent with momentum, AdaGrad, RMSprop, Adam Opti ion methods, Normalisation- Standardisation and min-max Normalization, Batch Normalization and Layer Normalization, Regularization- L1 and L2 Regularization, Dropout, and Data Augmentation, Early Stopping, Weight Decay, Hyperparameter Tuning. Convolutional Neural Networks Building Blocks- Forward Propagation in CNN, Backpropagation in Convolutional Layer and Forward and back propagation in Pooling layers, Seminal Architecture- BFF, ResNet and Inception Net Recurrent Neural Networks Introduction to RNN-Vanilla RNNs and Backpropagation through time (BPTT) in RNN, Computation for LSTM and Computation for GRU, CNN + RNN and CNN- LSTM Attention and Neural Computation Neural Attention-Encoder decoder architectures, Attention and its types, Neural Computation- Neural Computers, Neural Turing Machines and Differentiable Neural Computers Deep Unsupervised Learning ‘Taxonomy, Boltzmann Machines, Variational- AutoEncoders, Representation Learning, Variational AutoEncoders and Reparameterization Trick, Generative Adversarial Networks (GANs)- Challenges, Wassertein GAN, Deep ConvolutionalModule 6 Module 7 Module & Syllabus Deep Reinforcement Learning Continuous Problems- Introduction to Function approximation and Function Approximation, Value based Problems - Deep Q-Netweork and DON (Programming material), Policy based- Policy Optimization, Policy Gradient Theorem and Algorithm, Combining Policy & Value based Learning- Actor-Critic and Deep deterministic policy Gradient (DDPG) Applied Deep Learning Production Techniques- Serving API endpoints, Serving and Data Pipelines and Post Production, Federated Learning, Edge Computing- On Device Inference and Edge Computing, Model Quantization, Tensorflow Lite and Tensorfiow JS, Explainable Al Current Research In future] Graph Neural Network. Application of GNN, Meta Learning, Few shot and Zero shot learning, Bayesian optimization and Architecture searchfuse (ai Computer Vision Course Objectiv After the completion of this course, students will be able to: + Explore the field of computer vision from its very basics, starting with iat ce -B ole oleR- Ue Reece oR RU oe Relate to the challenges of Computer Vision and the evolution of ele e ar ae RR eile [om Pee Ue Et hs scr Tle uur ry Czy ee Ca eur geuka tages Extend basic programming knowledge while getting to work with handy open source CV tools for image manipulation tasks.fuse (ai Syllabus Module 1 Introduction to Course Introduction to Computer Vision and its applications, Image Manipulation Python Tools- OpenC\, Pillow tool, scikit-image tools, Numpy tools, TensorFlow image tools. Module 2 Image Formation and Representation Imaging- Lambertian model for image formation, Camera and Lens, Image as a function, Image Representation- Image as a matrix, Image as a 3D tensor, Color Spaces, Imaging Geometry, Module 3 Image Processing Point Operators-Simple Point Operators, Histogram Equalization, Binary Operators, Masks, Image Thresholding, Filters, Cross-Correlation and Convolution, Application of Filters, Denoising- Salt and Pepper noise, Gaussian Smoothing and Median Filtering, Sharpening, Out of focus blur (Gaussian Blur), Motion blur (box filter, Image Upscaling and Downscaling, Morphological Processing- Erosion and Dilation, Closing and Opening Module 4 Feature Detection and Matching Edge Detectionintroduction to Edge Detection- 1st order and 2 order edge detection techniques, Canny Edge Detection, RANdom SAmple Consensus (RANSAC), Feature Detection and Matching- Local Invariant Features and Keypoint Localization Module 5 Segmentation Introduction and types of segmentation, Split and Merge Watershed, Mode finding Segmentation, K-means Clustering, Mean Shift Clustering Module6 = Object Recognition Introduction to Recognition task, Classification problem in Image, Localization, Template Matching, Detection problem, Sliding Window and PyramidsModule 7 Module 8 Syllabus Video Processing Introduction to Video Processing, Video Formation, Video manipulation with ‘OpenCV, Tracking Problem - Color Based Tracking, MeanShift Tracking, Action Recognition and Gesture Recognition Deep Learning in Computer Vision Introduction to CNN, Building Blocks of CNN, Classification- LeNet and Few architectures for classification, Transfer Learning, Localization CNN based Localization, Detection- CNN Based Detectors and Advanced Object Detectors, Segmentation- Segmentation using FCN, Unsupervised Methods- Auto Encodersfuse (ai (D Natural Language Processing Course Objectiv After the completion of this course, students will be able to: Cena aa eee teeta nL Cem eee com LoL) intuition on the possibilities of real world applications. Create basic NLP pipelines using common tools and operations to rod aC cn Summarize the architectures of neural language models, continuous bag of words model and skip gram model to train word embeddings. Define different linguistic terminologies (morphology, tagging, syntax, parsing, semantics, pragmatics and discourse), morphological classes and derivation phenomenons to understand Nie are EU noetse cama alle ce stsfuse |ai Syllabus Introduction to Natural Language Processing Introduction to NLP, Application of NLP, Introduction to NLP Tools, loading corpus and lexicon other than one loaded in the reading material. Language and Knowledge Introduction to Language and knowledge, Morphology, Tagging, Syntax and Parsing, Context Free Grammar, Parsing, Semantics, Pragmatics and Discourse, Lexical Semantics and WordNet, Word Similarity and Vector Semantics and Pragmatics and Discourse Machine Learning for NLP Basic NLP Pipeline, Raw Text Processing- Loading Data from from different sources, Basic Python string processing, Regular Expression and Common UNIX tools basics, Text Processing Text Cleaning and Tokenization, Normalization and. Stopwords Removal, Vectorization Machine Learning, Applying Machine Learning Language Models Markov Models- Introduction to Markov Assumption, N-Gram Language Model, Evaluation of Language Models, Generalization and Zeros, Smoothing and Backoft, Word Embedding-Introduction to Word Embeddings- Semantic Interpretations and Properties, Neural Language Models, Word2Vec and Glove Embeddings Deep Learning for NLP ‘Sequence Models- RNNs for NLP, CNN for NLP and Encoder-Decoder, Attention, Contextual Embedding and Pretraining